SUPIR is a large-scale image restoration model developed by the XPixel Group. Built on a diffusion-based generative approach, it can recover structure and fine details even from heavily degraded images.
Text-guided restoration
SUPIR supports text prompts to steer the restoration result. This is useful when you need more than denoising and artifact removal—for example, adjusting the look or overall mood of a scene while restoring it.
High-fidelity restoration and upscaling
The model focuses on preserving realistic details while improving low-quality photos, landscapes, and other images. It increases resolution, reduces distortions, and aims to reconstruct textures without turning the output into an overly smooth, “plastic” render.
For research and production workflows
SUPIR is available via an online app and is also accompanied by a GitHub repository and a research paper. This makes it practical for real-world image enhancement tasks and useful for researchers working on modern diffusion-based restoration methods.
